What are the hardware requirements for GPT Sw3 20B?

GPT Sw3 20B requires approximately 83.2 GB of VRAM. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed GPT Sw3 20B and how large is it?

GPT Sw3 20B is developed by AI-Sweden-Models, a model with 20b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does GPT Sw3 20B perform on standard benchmarks?

GPT Sw3 20B has the following published scores: MMLU 28.47. Compare against reference models on this page or on the LLM Explorer leaderboards.

GPT Sw3 20B by AI-Sweden-Models — VRAM 83.2GB

Name: GPT Sw3 20B
Author: AI-Sweden-Models

GPT Sw3 20B is an open-source language model by AI-Sweden-Models. Features: 20b LLM, VRAM: 83.2GB, License: other, LLM Explorer Score: 0.09, Arc: 41.8, HellaSwag: 68.8, MMLU: 28.5, GSM8K: 1.

Da En Endpoints compatible Gpt2 Is No Pytorch Region:us Safetensors Sharded Sv Tensorflow

Model Card on HF 🤗: https://huggingface.co/AI-Sweden-Models/gpt-sw3-20b

GPT Sw3 20B Parameters and Internals

Model Type

decoder-only, transformer, language model

Use Cases

Areas:

research, evaluation of LLM capabilities

Applications:

Nordic NLP ecosystem validation and testing

Primary Use Cases:

language generation tasks

Limitations:

bias, safety, generation diversity, hallucination

Additional Notes

GPT-SW3 is compliant with the Modified RAIL license focusing on communication and transparency regarding LLMs.

Supported Languages

da (advanced), sv (advanced), no (advanced), en (advanced), is (advanced)

Training Details

Data Sources:

Books, Articles, Code, Conversational, Math, Miscellaneous, Web Sources

Data Volume:

320B tokens

Methodology:

NeMo Megatron GPT implementation

Model Architecture:

decoder-only transformer

Release Notes

Version:

Second Generation

Date:

2022-12-20

Notes:

Release of second generation of GPT-SW3 model.

LLM Name	GPT Sw3 20B
Repository 🤗	https://huggingface.co/AI-Sweden-Models/gpt-sw3-20b
Model Size	20b
Required VRAM	83.2 GB
Updated	2026-05-30
Maintainer	AI-Sweden-Models
Model Type	gpt2
Model Files	9.5 GB: 1-of-9 9.7 GB: 2-of-9 9.7 GB: 3-of-9 9.7 GB: 4-of-9 9.7 GB: 5-of-9 9.7 GB: 6-of-9 9.7 GB: 7-of-9 9.7 GB: 8-of-9 5.8 GB: 9-of-9 9.5 GB: 1-of-9 9.7 GB: 2-of-9 9.7 GB: 3-of-9 9.7 GB: 4-of-9 9.7 GB: 5-of-9 9.7 GB: 6-of-9 9.7 GB: 7-of-9 9.7 GB: 8-of-9 5.8 GB: 9-of-9
Supported Languages	da sv no en is
Model Architecture	GPT2LMHeadModel
License	other
Transformers Version	4.25.0.dev0
Tokenizer Class	GPTSw3Tokenizer
Vocabulary Size	64000
Torch Data Type	float32
Activation Function	gelu

Rank the GPT Sw3 20B capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to GPT Sw3 20B

Best Alternatives	Context / RAM	Downloads	Likes
Palmyra 20B Chat	0K / 40.6 GB	173	11
Palmyra Med 20B	0K / 40.6 GB	224	37
GPT Sw3 20B Instruct	0K / 83.2 GB	1129	12
InstructPalmyra 20B	0K / 40.6 GB	158	40

Note: green Score (e.g. "73.2") means that the model is better than AI-Sweden-Models/gpt-sw3-20b.